4 #include <linux/blkdev.h>
5 #include <linux/sbitmap.h>
6 #include <linux/srcu.h>
9 struct blk_flush_queue;
11 struct blk_mq_hw_ctx {
14 struct list_head dispatch;
15 unsigned long state; /* BLK_MQ_S_* flags */
16 } ____cacheline_aligned_in_smp;
18 struct work_struct run_work;
19 cpumask_var_t cpumask;
23 unsigned long flags; /* BLK_MQ_F_* flags */
26 struct request_queue *queue;
27 struct blk_flush_queue *fq;
31 struct sbitmap ctx_map;
33 struct blk_mq_ctx **ctxs;
38 struct blk_mq_tags *tags;
39 struct blk_mq_tags *sched_tags;
41 struct srcu_struct queue_rq_srcu;
45 #define BLK_MQ_MAX_DISPATCH_ORDER 7
46 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
48 unsigned int numa_node;
49 unsigned int queue_num;
53 struct delayed_work delay_work;
55 struct hlist_node cpuhp_dead;
58 unsigned long poll_considered;
59 unsigned long poll_invoked;
60 unsigned long poll_success;
63 struct blk_mq_tag_set {
65 const struct blk_mq_ops *ops;
66 unsigned int nr_hw_queues;
67 unsigned int queue_depth; /* max hw supported */
68 unsigned int reserved_tags;
69 unsigned int cmd_size; /* per-request extra data */
72 unsigned int flags; /* BLK_MQ_F_* */
75 struct blk_mq_tags **tags;
77 struct mutex tag_list_lock;
78 struct list_head tag_list;
81 struct blk_mq_queue_data {
83 struct list_head *list;
87 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
88 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
89 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
90 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
91 typedef int (init_request_fn)(void *, struct request *, unsigned int,
92 unsigned int, unsigned int);
93 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
95 typedef int (reinit_request_fn)(void *, struct request *);
97 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
99 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
100 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
101 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
108 queue_rq_fn *queue_rq;
111 * Called on request timeout
116 * Called to poll for completion of a specific tag.
120 softirq_done_fn *complete;
123 * Called when the block layer side of a hardware queue has been
124 * set up, allowing the driver to allocate/init matching structures.
125 * Ditto for exit/teardown.
127 init_hctx_fn *init_hctx;
128 exit_hctx_fn *exit_hctx;
131 * Called for every command allocated by the block layer to allow
132 * the driver to set up driver specific data.
134 * Tag greater than or equal to queue_depth is for setting up
137 * Ditto for exit/teardown.
139 init_request_fn *init_request;
140 exit_request_fn *exit_request;
141 reinit_request_fn *reinit_request;
143 map_queues_fn *map_queues;
147 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
148 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
149 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
151 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
152 BLK_MQ_F_TAG_SHARED = 1 << 1,
153 BLK_MQ_F_SG_MERGE = 1 << 2,
154 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
155 BLK_MQ_F_BLOCKING = 1 << 5,
156 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
157 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
159 BLK_MQ_S_STOPPED = 0,
160 BLK_MQ_S_TAG_ACTIVE = 1,
161 BLK_MQ_S_SCHED_RESTART = 2,
163 BLK_MQ_MAX_DEPTH = 10240,
165 BLK_MQ_CPU_WORK_BATCH = 8,
167 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
168 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
169 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
170 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
171 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
172 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
174 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
175 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
176 struct request_queue *q);
177 int blk_mq_register_dev(struct device *, struct request_queue *);
178 void blk_mq_unregister_dev(struct device *, struct request_queue *);
180 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
181 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
183 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
185 void blk_mq_free_request(struct request *rq);
186 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
189 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
190 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
191 BLK_MQ_REQ_INTERNAL = (1 << 2), /* allocate internal/sched tag */
194 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
196 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
197 unsigned int flags, unsigned int hctx_idx);
198 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
201 BLK_MQ_UNIQUE_TAG_BITS = 16,
202 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
205 u32 blk_mq_unique_tag(struct request *rq);
207 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
209 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
212 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
214 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
218 int blk_mq_request_started(struct request *rq);
219 void blk_mq_start_request(struct request *rq);
220 void blk_mq_end_request(struct request *rq, int error);
221 void __blk_mq_end_request(struct request *rq, int error);
223 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
224 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
225 bool kick_requeue_list);
226 void blk_mq_kick_requeue_list(struct request_queue *q);
227 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
228 void blk_mq_abort_requeue_list(struct request_queue *q);
229 void blk_mq_complete_request(struct request *rq, int error);
231 bool blk_mq_queue_stopped(struct request_queue *q);
232 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
233 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
234 void blk_mq_stop_hw_queues(struct request_queue *q);
235 void blk_mq_start_hw_queues(struct request_queue *q);
236 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
237 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
238 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
239 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
240 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
241 busy_tag_iter_fn *fn, void *priv);
242 void blk_mq_freeze_queue(struct request_queue *q);
243 void blk_mq_unfreeze_queue(struct request_queue *q);
244 void blk_mq_freeze_queue_start(struct request_queue *q);
245 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
247 int blk_mq_map_queues(struct blk_mq_tag_set *set);
248 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
251 * Driver command data is immediately after the request. So subtract request
252 * size to get back to the original request, add request size to get the PDU.
254 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
256 return pdu - sizeof(struct request);
258 static inline void *blk_mq_rq_to_pdu(struct request *rq)
263 #define queue_for_each_hw_ctx(q, hctx, i) \
264 for ((i) = 0; (i) < (q)->nr_hw_queues && \
265 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
267 #define hctx_for_each_ctx(hctx, ctx, i) \
268 for ((i) = 0; (i) < (hctx)->nr_ctx && \
269 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)